Condition from atkindon's paper http://info.asprs.org/publications/pers/2005journal/july/2005_july_839-846.pdf
for every sub-pixel the attractiveness Ai of a pixeli is predicted as a distance-weighted function of its j 1,2,. . . , J neighbors:
A(i)= sum (j=1 to J) { lambda(i,j)*z(x(j))}where z(x(j)) is the (binary) class of the jth pixel at location x(j),and lambda(i,j) is a distance-dependent weight predicted as:
lambda(i,j)= exp(-h(i,j)/a)where h(i,j) is the distance between the location x(i) of pixel ifor which the attractiveness is desired, the location x(j) of aneighboring pixel j, and "a" is the non-linear parameter of theexponential model.
My question is how to calculate h(i,j) and what is going to be the input for x(i) and x(j), the things I know as input are the class labels of the subpixels but how to calculate distance using them.
As in my case x(i) can be 1 as class label and x(j) can be 2. till now I am able to randomly fill the subpixel classes in a finer grid.
أكثر...
for every sub-pixel the attractiveness Ai of a pixeli is predicted as a distance-weighted function of its j 1,2,. . . , J neighbors:
A(i)= sum (j=1 to J) { lambda(i,j)*z(x(j))}where z(x(j)) is the (binary) class of the jth pixel at location x(j),and lambda(i,j) is a distance-dependent weight predicted as:
lambda(i,j)= exp(-h(i,j)/a)where h(i,j) is the distance between the location x(i) of pixel ifor which the attractiveness is desired, the location x(j) of aneighboring pixel j, and "a" is the non-linear parameter of theexponential model.
My question is how to calculate h(i,j) and what is going to be the input for x(i) and x(j), the things I know as input are the class labels of the subpixels but how to calculate distance using them.
As in my case x(i) can be 1 as class label and x(j) can be 2. till now I am able to randomly fill the subpixel classes in a finer grid.
أكثر...